Ecology, 96(7), 2015, pp. 1943–1956� 2015 by the Ecological Society of America
Natural selection by pulsed predation: survival of the thickest
ALLERT I. BIJLEVELD,1,4 SONKE TWIETMEYER,1,2 JULIA PIECHOCKI,1 JAN A. VAN GILS,1 AND THEUNIS PIERSMA1,3
1Department of Marine Ecology, NIOZ Royal Netherlands Institute for Sea Research, P.O. Box 59, 1790AB Den Burg,The Netherlands
2Department of Biogeography, Trier University, Universitatsring 15, D-54296 Trier, Germany3Animal Ecology Group, Groningen Institute for Evolutionary Life Sciences, University of Groningen, P.O. Box 11103,
9700CC Groningen, The Netherlands
Abstract. Selective predation can lead to natural selection in prey populations and mayalleviate competition among surviving individuals. The processes of selection and competitioncan have substantial effects on prey population dynamics, but are rarely studiedsimultaneously. Moreover, field studies of predator-induced short-term selection pressureson prey populations are scarce. Here we report measurements of density dependence in bodycomposition in a bivalve prey (edible cockle, Cerastoderma edule) during bouts of intensepredation by an avian predator (Red Knot, Calidris canutus). We measured densities,patchiness, morphology, and body composition (shell and flesh mass) of cockles in a quasi-experimental setting, i.e., before and after predation in three similar plots of 1 ha each, two ofwhich experienced predation, and one of which remained unvisited in the course of the shortstudy period and served as a reference. An individual’s shell and flesh mass declined withcockle density (negative density dependence). Before predation, cockles were patchilydistributed. After predation, during which densities were reduced by 78% (from 232 to 50cockles/m2), the patchiness was substantially reduced, i.e., the spatial distribution washomogenized. Red Knots selected juvenile cockles with an average length of 6.9 6 1.0 mm(mean 6 SD). Cockles surviving predation had heavier shells than before predation (anincrease of 21.5 percentage points), but similar flesh masses. By contrast, in the reference plotshell mass did not differ statistically between initial and final sampling occasions, while fleshmass was larger (an increase of 13.2 percentage points). In this field study, we show that RedKnots imposed a strong selection pressure on cockles to grow fast with thick shells and littleflesh mass, with selection gradients among the highest reported in the literature.
Key words: bivalves; competitive release; density dependence; directional selection; foraging; intertidalmudflats; natural selection; phenotypic selection; predator defense; selective predation; shorebirds; soft-sediment habitat.
INTRODUCTION
Predation is a key process in the ecology and
population dynamics of prey (Vermeij 1987, Krebs and
Davies 1997), and selective predation is an important
agent of natural selection due to the removal of specific
classes of prey (Endler 1986, Wade and Kalisz 1990),
leading to the evolution of predator defense mechanisms
(Caro 2005). Furthermore, when prey suffer from
density-dependent processes, by thinning prey densities,
predation can alleviate competition between surviving
individuals (de Roos and Persson 2013). For instance, if
individual growth is negatively density dependent,
predation reduces competition and thus enhances the
growth of surviving individuals. Predation can thus have
a major influence on densities, patchiness, size structure,
body composition, and potentially the reproductive
output of prey at the population level (Gurevitch et al.
2000, de Roos and Persson 2013).
Although predator–prey interactions have been stud-
ied for a long time (Verhulst 1838, Krebs and Davies
1997, Caro 2005), studies that quantify the short-term
selection pressures by predators on prey populations are
rare, especially in the wild (Endler 1986, Calsbeek and
Cox 2010). Here, we report on natural predation by Red
Knots Calidris canutus on edible cockles Cerastoderma
edule burrowed at shallow depths in temperate intertidal
mudflats without experimental artefacts resulting from
predator exclosures. In fact, we utilized the spatial
unpredictability in the occurrence of flocks of foraging
Red Knots (Folmer et al. 2010) to provide us with
predation plots as well a reference plot without
predation.
We quantified densities, patchiness, and external
morphology (shell length, width, and height), as well
as body composition (shell and flesh mass) of cockles in
their natural environment. We were able to quantify
these variables before and after a two-week pulse of
intense predation by Red Knots, as well as in a situation
Manuscript received 7 October 2014; revised 7 January 2015;accepted 12 January 2015. Corresponding Editor: G. J.Vermeij.
4 E-mail: [email protected]
1943
without predation; the latter served as a reference. The
comparison of cockles between the predation and the
reference plots enabled us to study effects of predators
on their prey in this quasi-experimental natural setting.
Note that we consider it a quasi-experiment because we
did not control where the birds foraged. We will show
that the cockles suffered from intraspecific competition
(negative density dependence), and that Red Knots can
have profound effects on the length distribution of
cockles, as well as their density, patchiness, and body
composition. We calculated selection gradients that were
among the highest reported in the literature, and showed
that Red Knots imposed strong selection on cockles to
grow fast with thick shells and little flesh mass.
METHODS
Background
Red Knots (see Plate 1) are medium-sized shorebirds
that during the nonbreeding season live in tidal areas
(Piersma 2007, 2012). They are social and foraging
groups of up to several thousand individuals are
common (Piersma et al. 1993). Over short timescales
(weeks) their foraging locations tend to be unpredict-
able, which is attributed to their strong social attraction
(Folmer et al. 2010), mobility (van Gils et al. 2005b), and
the large spatial extent of foraging opportunities (Kraan
et al. 2009). Within each low-tide period, Red Knots fly
tens of kilometers across exposed mudflats in search of
buried hard-shelled mollusks such as edible cockles
(Piersma et al. 1993, van Gils et al. 2005b). Because they
swallow their prey whole, Red Knots are limited to
ingesting cockles smaller than 16 mm (Zwarts and
Blomert 1992, Piersma et al. 1993) and constrained by
the amount of shell material that they can process (van
Gils et al. 2003a). Due to this digestive constraint, Red
Knots maximize their energy intake rates by selecting
individual cockles with large flesh mass compared to
their shell mass (van Gils et al. 2005a).
Cockles are suspension-feeding bivalves that are
commonly found in the Dutch Wadden Sea (Beukema
et al. 1993). Their spatial distribution is widespread
(Kraan et al. 2009), and they can be found in densities of
up to several thousand individuals/m2 (Jensen 1993).
Cockle population size, as well as recruitment, greatly
varies between years (Beukema et al. 1993). They spawn
between May and August, leading to distinct year
classes (Beukema et al. 2001). After a planktonic phase
of several weeks, they settle on mudflats when they are
;0.3 mm long (e.g., De Montaudouin and Bachelet
1996). Cockles live in mudflats with inundation times
ranging from 2 to 12 hours and sediment grain sizes
ranging from 75 to 275 lm (Kraan et al. 2010).
Nonetheless, cockles prefer mudflats with inundation
times between 6 and 8 hours (Kraan et al. 2010). Due to
PLATE 1. A group of foraging Red Knots (Calidris canutus islandica) on intertidal mudflats. Some of the birds have found ediblecockles (Cerastoderma edule) that they are about to swallow whole. Photo credit: Jan van de Kam.
ALLERT I. BIJLEVELD ET AL.1944 Ecology, Vol. 96, No. 7
short siphons, they are limited in their burying depth
and are found within a few centimeters of the surface
(Zwarts and Wanink 1989). Living close to the surface
and within reach of predators (e.g., Red Knots with
their 4 cm long bills), cockles rely on predator defenses
such as shell thickness (armor). Cockles can grow to a
maximum of 50 mm, and generally don’t live longer
than 5 years (Beukema et al. 1993).
Study design
Our study site was located in the Dutch Wadden Sea
on the tidal flats near the uninhabited islet of Griend
(53814.6150 N, 5815.2190 E; Appendix: Fig. A1). Griend
is surrounded by extensive intertidal mudflats that
stretch for tens of kilometers. Near Griend, we selected
three plots (plots A, B, and C) of 100 3 100 m each
where Red Knots were previously seen foraging on
cockles. All plots were visually identical, located at
similar distances from Griend (590, 660, and 520 m for
plots A, B, and C, respectively), and had similar
inundation times (7.6, 7.7, and 6.7 hours for plots A,
B, and C, respectively) and sediment grain sizes (182,
182, and 185 lm for plots A, B, and C, respectively; see
Compton et al. [2013]). Given the wide range of
inundation times (from 2 to 12 hours) and sediment
structures (from 75 to 275 lm) that cockles occur at
(Kraan et al. 2010), the differences in habitat charac-
teristics between the plots are actually small. In fact, all
plots fall within the preferred habitat range of cockles
(Kraan et al. 2010). Due to difficulty in predicting where
Red Knots would forage within a tide (Folmer et al.
2010), we did not know beforehand at which plot, if any,
Red Knots would forage. Two of the three plots were
visited by knots (plots A and B), and even though we
had seen foraging knots there as well, plot C was not
visited by Red Knots during our measurement interval.
This allowed us to study the effect of Red Knot
predation on cockles in comparison to a reference plot
without predation, i.e., a before–after control–impact
design. All three plots were sampled intensively over a
relatively large spatial scale (1 ha).
Sampling the prey
On 12 and 18 August and on 4 September 2010 we
sampled cockle densities in plots A, B, and C,
respectively. On 26 August, Red Knots gave up foraging
in plots A and B. On 26 August, 2 and 9 September, we
resampled cockle densities in plots A, B, and C,
respectively. Although ideally we should have sampled
all plots simultaneously, logistical limitations prevented
us from doing so. Nonetheless, the sampling dates were
relatively close together and we have no reason to
suspect that factors that vary over time have influenced
our results (Zwarts 1991). At each plot we sampled 150
stations, of which 100 sampling stations were placed 10
m apart on a systematic grid, and the remaining 50
sampling stations were randomly placed on grid lines
(Bijleveld et al. 2012). This sampling design allowed for
precise estimation of mean densities, as well as spatial
autocorrelation parameters that were necessary for
estimating patchiness and accurate spatial interpolations
of cockle densities (Bijleveld et al. 2012).
We marked sampling stations with color-coded PVC
tubes (20 mm in diameter) reaching 20 cm above the
mudflat. We avoided resampling the exact locations by
initially sampling east and finally sampling 10 cm west of
the marker. At each sampling station we collected one
core (1/56 m2) to a depth of 20–25 cm, which we rinsed
over a 1-mm mesh sieve. We collected and froze all
cockles before taking them to the laboratory, where
their lengths, widths, and heights (as defined by Zwarts
and Blomert 1992) were measured to the nearest 0.1 mm.
From a subsample of 115 sampling stations (1094
individuals), we determined an individual’s body com-
position by measuring dry mass of the shell (DMshell)
and ash-free dry mass of the flesh (AFDMflesh)
according to the procedure described by Piersma et al.
(1993). These body compositional samples were uneven-
ly distributed between plots and sampling occasions. For
the first and second sampling occasion we sampled 21
and 0 individuals from plot A, 186 and 72 individuals
from plot B, and 214 and 601 individuals from plot C. In
order to correct for this uneven distribution, we
analyzed the data in mixed-effect analyses with sampling
station as random effect (see Data analyses).
Spatially autocorrelated cockle densities
Often, animal densities are positively correlated over
small distances, and the farther apart, the weaker this
correlation. A spatial autocorrelation function describes
how spatial autocorrelation changes with distance, and
can be used for estimating the average patch size (e.g.,
Kraan et al. 2009), or for spatial interpolations (Cressie
1993). In order to investigate the effect of predation on
the patchiness of cockles, as well as to visualize their
spatial distributions, we calculated spatial autocorrela-
tion functions and interpolated cockle densities across
each plot. Per sampling core, we counted the number of
cockles that were suitable for Red Knots to swallow
(,16 mm). We normalized model residuals by trans-
forming the numbers of suitable cockles with the
common logarithm (log10). To avoid taking the loga-
rithm of zero, we added one before the data transfor-
mation. We calculated a correlogram based on the
(transformed) numbers of suitable cockles per sampling
core for each plot with a spatial lag of 3 m. We then
fitted several commonly used spatial autocorrelation
functions to the correlograms and selected the exponen-
tial spatial autocorrelation function (van der Meer and
Leopold 1995, Bijleveld et al. 2012) that had the lowest
Akaike Information Criterion between all plots.
In order to estimate the average cockle patch size
within plots, we calculated at what distance the
autocorrelation reached the arbitrary value of 0.1
(Kraan et al. 2009). In the presence of spatial
autocorrelation, we estimated mean cockle densities
July 2015 1945NATURAL SELECTION BY PULSED PREDATION
and their standard errors with generalized least squares
(Cliff and Ord 1981); otherwise we used ordinary least
squares analyses (Bijleveld et al. 2012). For each plot, we
spatially interpolated cockle densities with kriging
(Cressie 1993). For representation purposes, we back-
transformed the density estimates with their 95%confidence interval (CI) and divided these by the surface
area of the sampling core to obtain cockle densities in
numbers per square meter. Note that as a result of the
logarithmic transformations, model results represent
geometric means instead of arithmetic means. In order
to correct for this bias and obtain the arithmetic means,
we multiplied the back-transformed estimates by the
antilog of 0.5 3 loge(10) 3 r2 (Rothery 1988).
Sampling predator abundances
In order to estimate densities of foraging Red Knots
in the study plots, we video-recorded each plot, in good
weather during the daytime, during low tide for as long
the plot was studied (between the initial and final prey
sampling of each plot). By slowly moving the camera
from left to right, each plot could be captured entirely by
one camera. In total we video-recorded Red Knots for
15 and 22 hours in plots A and B, and for 0 hours in plot
C, as there were no Red Knots present during the short
study period. Based on these recordings we estimated
that an average of 74 6 4.9 Red Knots (mean 6 SD) per
plot were present in plots A and B for an average
duration of 2 hours per tide, and none in plot C.
Data analyses
Due to nonlinearity and heteroscedasticity, the
allometric relationships between body composition and
length are usually analyzed with linear regressions on a
log-log scale. However, due to remaining nonlinearity,
we modelled an individual’s DMshell and AFDMflesh
with length on a log-log scale using nonlinear local
regression models (LOESS with local quadratic fitting;
Appendix: Fig. A2) (for R-script see Bijleveld et al.
2015). LOESS is flexible and follows the data regardless
of any nonlinear patterns. To compare DMshell and
AFDMflesh between differently sized cockles, we extract-
ed an individual’s residual from the nonlinear LOESS
fits, which reflects their relative DMshell and AFDMflesh.
For representation purposes, we back-transformed these
residuals into ratios representing an individual’s body
composition relative to the expected value for that
length. Note that even though shell length is a one-
dimensional measure of body size, our results were
similar to analyses with three-dimensional measures of
size (length 3 width 3 height). Because length is a more
intuitive measure of size than the three-dimensional
multiplication and has been used in bivalve studies
before (e.g., Armonies 1992, Zwarts and Blomert 1992,
Piersma et al. 1995, van Gils et al. 2005b), all our
analyses are based on length. In order to select the
smoothing parameter of the LOESS fits, we inspected
the pattern of model residuals with length. A smoothing
parameter of 0.5 gave the smoothest fits (i.e., removed
the size dependence) while still following the structural
features of our data (Jacoby 2000). In order to assess the
goodness-of-fit, resembling the coefficient of determina-
tion r2, we calculated the ratio of the sum of squares in
the LOESS fitted values to the total sum of squares in
the dependent variable (Jacoby 2000). The calculated
ratios for DMshell and AFDMflesh were 0.99 and 0.98,
but note that, in comparison to r2 values, the
interpretation of these ratios is not straightforward
(Jacoby 2000).
For the density dependence analyses we included
cockles from all plots, but excluded those samples from
the final sampling occasions in the predation plots.
Density dependence is a result of intraspecific competi-
tion that is not limited to specific size classes, i.e., size
classes that Red Knots can swallow. We therefore
included cockles of all lengths (between 3.6 and 41.6
mm) in the analyses of density dependence. With this
subset of data we calculated an individual’s relative
DMshell and AFDMflesh as described previously, and
analyzed these traits in linear mixed-effect models with
sampling station as a random effect, and shell length (in
millimeters) plus log10-transformed cockle density (in
square meters), and their interaction, as explanatory
variables. A significant interaction between length and
density on an individual’s relative DMshell or AFDMflesh
would indicate that cockles of different lengths are
differentially affected by density dependence (intraspe-
cific competition). In order to avoid computational
issues due to collinearity between predictors, we
centered length and log10-transformed density by
subtracting their means (12.9 mm and 3.07, respective-
ly). By parametric bootstrapping (n ¼ 1000), we
calculated significance under the null hypothesis that
the estimated coefficients are zero.
To analyze the effects of knot predation on an
individual’s relative DMshell and AFDMflesh, we selected
cockles from all plots and sampling occasions, but only
those of suitable sizes for Red Knots to swallow (length
,16 mm, n¼887). With this subset of data we calculated
an individual’s relative DMshell and AFDMflesh as
previously described, and analyzed these traits in linear
mixed-effect models with sampling station as a random
effect, and ‘‘sampling occasion’’ (a factor coding for
either initial or final sampling) plus ‘‘predation’’ (a factor
coding for either the predation or reference plots), and
their interaction, as explanatory variables. Due to the
positive correlation between an individual’s relative
AFDMflesh and DMshell (r ¼ 0.29, P , 0.01), we also
analyzed these data in bivariate mixed-effect models, i.e.,
a model with AFDMflesh and DMshell simultaneously as
response variables. These results were, nevertheless,
similar to univariate analyses, and for brevity we present
the univariate mixed-effect models. We additionally
investigated the effect of predation on the shape of
cockle shells by calculating the ratio of both shell height
and shell width to length. We analyzed these ratios in
ALLERT I. BIJLEVELD ET AL.1946 Ecology, Vol. 96, No. 7
mixed-effect models as explained above, but included
cockle length (centered by subtracting its mean) as an
explanatory variable in the analyses of the ratio of width
to length to correct for its linear increase with shell
length (0.008 6 0.001 mm�1, P , 0.01). By parametric
bootstrapping (n ¼ 1000), we calculated significance
under the null hypothesis that the estimated coefficients
are zero.
We calculated linear and nonlinear selection gradients
(Lande and Arnold 1983, Endler 1986) on length and
body composition with multivariate models following
Johnson et al. (2012). As collinearity between variables
can make these multivariate analyses unreliable, we
calculated selection gradients for length in a multivariate
model with an individual’s relative AFDMflesh and
DMshell. The condition numbers of the resulting
variance–covariance matrices indicated no problems
with collinearity for either the predation (j ¼ 2.4) or
the reference plot (j¼1.5). By parametric bootstrapping
(n ¼ 1000), we calculated standard deviations and
significance of the selection gradients under the null
hypothesis that they are zero.
All data analyses were carried out in R v3.1.0 (R
Development Core Team 2013) with the packages ncf
for calculating correlograms, fields for spatial interpo-
lations, and lme4 for mixed-effect model analyses.
RESULTS
Density dependence
A cockle’s relative shell mass (DMshell) and flesh mass
(AFDMflesh) declined with cockle density (Table 1 and
Fig. 1). The interaction between length and density on
relative DMshell was nonsignificant (Table 1A). For
relative AFDMflesh this interaction was significantly
positive (Table 1B), indicating that smaller cockles were
proportionally more affected by intraspecific competi-
tion than larger ones.
Predation and the patchiness of prey
Before predation the cockles (length ,16 mm) were
patchily distributed (Fig. 2A, C, and E). This was
evidenced by the significant positive autocorrelation at
distance zero (b0) and the decline of autocorrelation with
distance (b1) that we measured in plot A (b0 ¼ 0.47 6
0.05 [mean 6 SE], P , 0.01, and b1¼�0.05 6 0.01, P ,
0.01); plot B (b0¼ 0.54 6 0.12, P , 0.01, and b1¼�0.076 0.02, P , 0.01); and plot C (b0 ¼ 0.35 6 0.10, P ,
0.01, b1 ¼�0.05 6 0.02, P ¼ 0.03; Fig. 3A, C, and E).
The distance at which autocorrelation dropped below
0.1 (the average patch size) was 31 m for plot A, 25 m for
plot B, and 24 m for plot C. Initial cockle densities were
on average 186 cockles/m2 (95% CI [119; 278 cockles]) in
plot A, 277 cockles/m2 (95% CI [210; 362 cockles]) in
plot B, and 1230 cockles/m2 (95% CI [1037; 1457
cockles]) in plot C.
Red knot predation reduced cockle densities by 72%in plot A and 83% in plot B to 52 cockles/m2 (95% CI
[42; 62 cockles]) and 48 cockles/m2 (95% CI [38; 59
cockles]) respectively (Fig. 2A–D). After predation, the
patchiness in cockle densities was substantially reduced
(homogenized), as shown by the nonsignificant spatial
autocorrelation parameters after predation for both plot
A (b0¼ 0.13 60.22, P¼ 0.56, and b1¼ 0.29 6 0.63, P¼0.66) and plot B (b0¼�0.06 6 0.15, P¼ 0.68, and b1¼�0.07 6 0.23, P ¼ 0.77) (Fig. 3B, D). Compared to the
initial sampling, mean cockle density in the reference
plot was similar to the final sampling (1280 cockles/m2,
95% CI [1030; 1587 cockles], Fig. 2E, F). There were
some differences in the spatial density distribution
between the initial and final sampling in the reference
plot (Fig. 2E, F), but these probably reflect sampling
error. The autocorrelation parameters (b0¼ 0.41 6 0.14,
TABLE 1. Mixed-modelling results for the effects of cockle lengths and densities on their relativebody composition.
Response variables and random effects Predictors Estimates SE P
A) Relative DMshell intercept �0.000 0.004 1.00length �0.000 0.000 0.69density �0.031 0.011 ,0.01length 3 density �0.002 0.001 0.08
Sampling station 0.023 0.003 ,0.01Residual 0.063 0.001 ,0.01
B) Relative AFDMflesh intercept �0.002 0.008 0.82length 0.001 0.000 0.16density �0.057 0.018 ,0.01length 3 density 0.006 0.001 ,0.01
Sampling station 0.053 0.005 ,0.01Residual 0.063 0.001 ,0.01
Notes: The mixed-modelling results for the effects of cockle length (mm) and density (cockles/m2) on an individual’s relative (A) dry mass of the shell (DMshell), and (B) ash-free dry mass of theflesh (AFDMflesh). Cockle density was log10-transformed. In order to avoid computational issuesdue to collinearity, covariates were centered with their mean length (12.9 mm) and log10-transformed density (3.07). The random effect estimates refer to standard deviations. Note thatthese data included cockles of all lengths (3.6–41.6 mm) and excluded data from the final samplingoccasions in the predation plots.
July 2015 1947NATURAL SELECTION BY PULSED PREDATION
P , 0.01, b1 ¼�0.06 6 0.03, P ¼ 0.04), as well as the
average patch size (24 m), were similar to those at initial
sampling (Fig. 3E, F).
Selective predation and phenotypic traits of the prey
The differences in length distribution and body
composition of cockles, before and after predation,
were pronounced. Before predation, the mean length of
suitable cockles (length , 16 mm) in both plots A and B
was 7.4 6 2.4 mm [mean 6 SD]), whereas after
predation the mean length increased to 10.4 6 2.9
mm; Fig. 4A, B). Subtracting the frequency distributions
of suitable cockles before and after predation suggests
that Red Knots had selected cockles with a mean length
of 6.9 6 1.0 mm. The length distribution of suitable
cockles in the non-predation reference plot C did not
differ between the initial (10.9 6 1.8 mm) and final
sampling (11.1 6 1.8 mm, Fig. 4C).
Predation had no effect on the shape of cockle shells,
as neither did the ratio of width to length before
predation (0.65 6 0.07 [mean 6 SD]) differ from that
after predation (0.66 6 0.08), nor did the ratio of height
to length differ between before (0.90 6 0.05) and after
predation (0.91 6 0.07). In reference plot C, the ratio of
cockle width to length did differ significantly (0.01 6
0.005 [mean 6 SE], P¼ 0.01) between the initial (0.66 6
0.07 [mean 6 SD]) and final sampling (0.67 6 0.05
[mean 6 SD]), and the ratio of height to length differed
significantly (0.01 6 0.003 [mean 6 SE], P , 0.01)
between initial (0.89 6 0.05 [mean 6 SD]) and final
sampling (0.90 6 0.04 [mean 6 SD]). The changes in
shell shape between initial and final sampling in the
reference plot were small and similar to the predation
plots, as neither did the changes in height-to-length
ratios (�0.004 6 0.007 [mean 6 SE], P ¼ 0.53), nor the
changes in width-to-length ratios (0.02 6 0.010 [mean 6
SE], P¼ 0.14) differ significantly between the predation
and reference plots.
Individuals surviving predation had heavier shells, an
increase of 21.5 percentage points (95% CI [12.4; 31.2], P
, 0.01; Fig 5A and Appendix: Table A1), than before
predation, indicating that predation affected cockle shell
thickness. An individual’s relative AFDMflesh did not
differ between the initial and final measures (6.4
percentage points higher, 95% CI [�5.1; 19.1], P ¼0.26; Fig 5B and Appendix: Table A1). In reference plot
C, we observed no significant difference in an individ-
ual’s relative DMshell between initial and final sampling
(2.4 percentage points, 95% CI [�3.3; 8.4], P¼ 0.42; Fig
5A and Appendix: Table A1). However, we did observe
that an individual’s relative AFDMflesh was 13.2
percentage points larger in the final sample compared
to the initial sample (95% CI [2.6; 25.1], P¼ 0.02; Fig 5B
and Appendix: Table A1).
Selection gradients
In the predation plots, we observed positive linear
selection gradients on cockle length and relative DMshell,
but we did not find this for an individual’s relative
AFDMflesh (Table 2). The nonlinear (quadratic) selec-
tion gradients on length, and relative DMshell were also
significantly positive, and we found that natural
selection favored combinations of large lengths and
small relative AFDMflesh (Table 2). In the reference plot,
we did not find a significant linear selection gradient on
an individual’s relative DMshell, but those on length and
relative AFDMflesh were significantly positive (Table 2).
In addition, the nonlinear selection gradient on DMshell
was significantly positive, and natural selection favored
FIG. 1. Negative density dependence in body compositionof cockles. An individual’s relative (A) dry mass of the shell(DMshell) and (B) ash-free dry mass of the flesh (AFDMflesh)were plotted against cockle densities. For representationpurposes, we back-transformed relative body composition intoratios representing an individual’s body composition relative tothe expected value for that length. Note that these data includedcockles of all lengths (3.6–41.6 mm) and excluded data from thefinal sampling occasions in the predation plots. The slope of theregression between relative AFDMflesh and cockle density inpanel (B) decreased with cockle length as indicated by thesignificantly positive interaction between cockle length anddensity (Table 1). For representation purposes, the regressionpresented in panel (B) shows the decline in an individual’srelative AFDMflesh with cockle density for 6.9 mm long cockles(i.e., mean cockle length eaten by Red Knots).
ALLERT I. BIJLEVELD ET AL.1948 Ecology, Vol. 96, No. 7
FIG. 2. Spatial density distributions of suitable-sized cockles that are suitable for Red Knots to swallow (length , 16 mm).Rows represent the different plots (respectively plots A, B, and C), and the columns represent the sampling occasion, with the initialsampling shown on the left (panels A, C, and E), and the final sampling shown on the right (panels B, D, and F). The top two rows(panels A–D) show the plots where cockles were fed upon by Red Knots, and the third row (panels E–F) shows the reference plotwhere Red Knots were not observed foraging. For the spatial representation of final densities (panels B and D) we spatiallyinterpolated densities with the autocorrelation function estimated from the initial sampling. White dots show the sampling stationsand the colors represent cockle densities (cockles/m2). Note that the density scales differ between plots.
July 2015 1949NATURAL SELECTION BY PULSED PREDATION
combinations of large lengths and large relativeAFDMflesh
(Table 2).
The significantly positive linear selection gradients on
length and AFDMflesh in the reference plot indicated
growth between the initial and final sampling period. In
order to account for such growth and investigate the net
effect of predation on natural selection, we subtracted
the linear selection gradients of the reference plot from
those of the combined predation plots. These adjusted
selection gradients confirmed that predation generated a
positive selection gradient on cockle length, a positive
selection gradient on relative DMshell, but also revealed a
FIG. 3. Spatial autocorrelation functions of the transformed numbers of suitable cockles (length , 16 mm) per sampling core.The rows represent the different plots (respectively plots A, B, and C), and the columns represent the sampling occasion, with theinitial sampling shown on the left (panels A, C, and E), and the final sampling shown on the right (panels B, D, and F). Plots A andB were visited by foraging Red Knots, and plot C was a reference plot without Red Knot predation. The initial autocorrelationfunctions are given by: y¼ 0.47e�0.05x for plot A, y¼ 0.54e�0.07x for plot B, and y¼ 0.42e�0.06x for plot C. The final autocorrelationfunction for plot C was 0.41e�0.06x.
ALLERT I. BIJLEVELD ET AL.1950 Ecology, Vol. 96, No. 7
FIG. 4. Effects of predation on the length distribution of cockles. We present the length distributions of cockles at initial andfinal sampling for (A) predation plot A, (B) predation plot B, and for (C) the reference plot without predation. The vertical linesindicate the upper limit (16 mm) of cockles that Red Knots are able to swallow. When the initial number of cockles was smallerthan that of the final number, a short horizontal line indicates the height of the underlying bar. Note the different scales of the y-axis.
July 2015 1951NATURAL SELECTION BY PULSED PREDATION
negative selection gradient on relative AFDMflesh (Table
2).
DISCUSSION
The processes of selection and competition are rarely
studied together, and field studies of predator-induced
short-term selection pressures on prey populations are
scarce. In this quasi-experimental field study, we showed
that cockles suffered from intraspecific competition, and
that selective predation by Red Knots has profound
effects on the density, the patchiness, as well as the
length distribution and body composition of their cockle
prey. Red Knots ate small cockles with thin shells and
proportionally large flesh content, imposing a strong
selection pressure on cockles to grow fast and have thick
shells with little flesh mass. Before discussing the
ecological implications of our study, we will first address
possible caveats in our study design.
Study design and robustness of results
The nature of our field study suggested some
problems of sampling design. The predation and
reference plots were sampled in sequence (the reference
plot was sampled 2–3 weeks after the predation plots).
The difference in depletion between the predation and
reference plots could therefore be confounded by some
(environmental) variable that changed over time causing
differential natural mortality or emigration between the
predation and reference plots. We do not think this is a
realistic concern, as in August–September the environ-
mental conditions in the Wadden Sea, and indeed cockle
body condition, tend to be stable (Zwarts 1991). Parada
FIG. 5. Effects of predation on cockle body composition. We present an individual’s relative (A) dried shell mass (DMshell), and(B) ash-free dry mass of the flesh (AFDMflesh) at the initial and final sampling occasion and separated for the predation plots andreference plot. For representation purposes, we back-transformed relative body composition into ratios representing an individual’sbody composition relative to the expected value for that length. Note that these data were limited to cockles that Red Knots wereable to swallow (lengths , 16 mm); n¼ number of cockles. The boxes indicate the inter-quartile range, the horizontal lines in theboxes indicate the median, the whiskers indicate the 95% data range, and the data points indicate the ,5% data range. The circlesrepresent model estimates from the Appendix: Table A1, which are connected within the predation treatment (solid lines) andreference treatment (dotted lines). Note the logarithmic scale on the y-axis.
TABLE 2. Cockle selection gradients imposed by Red Knot predation.
Selection gradient Trait
Predation Reference Difference
Est. SE P Est. SE P Est. SE P
Linear (b) length 1.39 0.28 ,0.01 0.40 0.09 ,0.01 1.00 0.29 ,0.01DMshell 1.41 0.33 ,0.01 0.02 0.09 0.82 1.39 0.34 ,0.01AFDMflesh �0.41 0.26 0.12 0.65 0.10 ,0.01 �1.07 0.28 ,0.01
Nonlinear (c) length 3.48 1.11 ,0.01 0.28 0.20 0.16 3.21 1.13 ,0.01length 3 DMshell 0.97 1.25 0.44 �0.05 0.15 0.75 1.02 1.26 0.42length 3 AFDMflesh �2.11 0.96 0.03 0.64 0.19 ,0.01 �2.75 0.98 ,0.01DMshell 4.77 1.76 ,0.01 0.36 0.18 ,0.05 4.41 1.77 0.01DMshell 3 AFDMflesh �0.06 1.10 0.96 0.12 0.16 0.44 �0.18 1.11 0.87AFDMflesh 1.44 0.82 0.08 0.38 0.26 0.14 1.06 0.86 0.22
Notes: We estimated cockle selection gradients for the predation and reference plot, and we show their differences. The traitsrefer to a cockle’s length in millimeters, as well as the relative dry mass of the shell (DMshell), and relative ash-free dry mass of theflesh (AFDMflesh). Note that we limited these analyses to cockles that Red Knots could ingest (lengths , 16 mm).
ALLERT I. BIJLEVELD ET AL.1952 Ecology, Vol. 96, No. 7
and Molares (2008) estimated the natural mortality of
cockles at 0.01 cockles/d, which, in our study, would
translate into a density reduction of 7% over the course
of 14 days. Thus, natural mortality alone cannot explain
the observed density changes.
Also cockle emigration rates seem too low to account
for the observed density reduction. Only spat up to a size
of 3.5 mm is capable of migration in the water column
over large distances (Armonies 1992). Larger cockles are
capable of crawling over the surface at speeds of 0.6 cm/
d (Flach 1996), but speeds of 50 cm/d have also been
reported (Mouritsen 2004). Such speeds will correspond
to an average linear movement of 0.08 m, and 7 m at the
most, during our short study period. These distances fall
comfortably within the 1-ha scale of our plots. Like
natural mortality, emigration does not seem capable of
reducing cockle densities by 72–83%.
In fact, the numbers of Red Knots that we observed in
the predation plots are capable of causing the observed
depletion. In our plots, Red Knots foraged on average
for 2 hours per tide, and selected 6.9 mm long cockles
with an average of 1.9 mg AFDMflesh (Appendix: Fig.
A2). In order to maintain their energy balance, Red
Knots require an intake rate of 0.3 mg AFDMflesh/s
(Piersma et al. 1995). The average difference of 182
cockles/m2 before and after predation would thus be
capable of sustaining 69 Red Knots per tide throughout
our study period of 24 low tides. This estimate is similar
to the 74 6 4.9 Red Knots (mean 6 SD) that we
observed per tide, which shows that Red Knot predation
would indeed cause a depletion of 72–83%.
The absence of true replication of the reference plot
leads to the question whether this is a sufficient
reference. We argue that the large spatial spread (across
1 ha) of the samples taken within the reference plot
should be seen as replication. Nevertheless, there were
differences between the predation and reference plot,
e.g., cockle abundance, size distribution. In ideal
circumstances the two treatments should only differ in
predation level. For a field study like this, the habitat
differences (e.g., in inundation time, sediment structure)
between the predation and reference plot were actually
very small (Kraan et al. 2010, but see Methods). In fact,
the reference and predation plots were all in the
preferred habitat range of cockles (Kraan et al. 2010),
and all plots contained cockles of suitable sizes to Red
Knots. There is nothing to suggest that the differences in
depletion between the predation and reference plots
would be caused by something other than predation.
Moreover, the presence of foraging Red Knots in the
reference plot, before and after the experimental
observation period, indicated its potential suitability to
Red Knots.
The timing between resampling the predation and
reference plots was different (14 days for the predation
and 5 days for the reference plot). That exposure to
potential predation was smaller, does not make the
unvisited reference plot less of a valid reference for lack
of predation. Nonetheless, the difference in timing might
affect the comparison of selection gradients between the
predation and reference plots (i.e., the net selection
gradients, Table 2). We would argue, however, that our
results are robust to this difference in sampling interval
for the following reasons. As a consequence of the
shorter sampling interval, we underestimated the in-
crease in mean length in the reference plot and
consequently overestimated the net selection gradient
on length. However, the increase in length due to growth
(over 14 days) was small compared to the increase in
mean length due to the selective removal of small size
classes (Fig. 4A, B).
The selection gradient for relative flesh mass was also
robust to the difference in sampling interval; in fact, the
estimate is conservative. Since the selection gradient on
flesh mass in the reference plot would have been larger
when given more time, subtracting this from the
selection gradient in the predation plot would have
resulted in a stronger (more negative) net selection
gradient. Note that the selection gradients resulting from
predation are as expected based on Red Knot foraging
behavior (e.g., van Gils et al. 2003a, 2005a).
Density dependence in the prey
Predation can have profound influences on the
population dynamics of species, especially when popu-
lation processes are density dependent (Gurevitch et al.
2000). For example, by reducing prey numbers, preda-
tion can reduce competition and enhance growth (van
Gils et al. 2012). This has a major influence on size
structure, and potentially on reproductive output at the
population level (Beukema et al. 2001, de Roos and
Persson 2013). Initially there was debate on whether
bivalve suspension feeders, such as cockles, can show
negative density dependence, as they are hypothesized to
be less susceptible to intraspecific competition for
resources (Levinton 1972). However, later empirical
studies showed that suspension-feeding bivalves are
actually susceptible to competition for space and/or
for food at even quite low densities (Peterson and Andre
1980, Jensen 1993, Kamermans 1993). In particular,
cockle growth (De Montaudouin and Bachelet 1996),
flesh content (Sutherland 1982), reproductive success
(Beukema et al. 2001), and survival (Parada and
Molares 2008) have been shown to decrease with
increased cockle densities. Here, we additionally dem-
onstrate declines in the relative shell and flesh mass of
cockles with density. We also show that the smallest
cockles were most susceptible to intraspecific competi-
tion on flesh mass (as indicated by the significant
interaction between length and density on AFDMflesh
[Table 1B]).
Depletion of cockle densities and community effects
Predators may substantially impact the densities of
their prey. Over the course of four months, for instance,
Common Eiders Somateria mollissima consumed be-
July 2015 1953NATURAL SELECTION BY PULSED PREDATION
tween 48% and 69% of their bivalve prey in an area of
6.7 ha (Guillemette et al. 1996). In a study on Red
Knots, it was shown that during single low-tide periods
they were able to take 25% of the bivalve stock (in this
particular case represented by Mya arenaria) in small
areas (100 m2; van Gils et al. 2003b). Among the most
substantial prey depletion reported in literature is that of
a combination of different wader species foraging on
chironomid larvae in 100-m2 plots, which decreased in
density by 87% over the course of 13 days (Szekely and
Bamberger 1992).
Prey depletion is often studied by means of predator
exclosures (Sih et al. 1985), artifacts that in intertidal
soft-sediment systems tend to alter the physical envi-
ronment and affect prey behavior, growth, etc. (Wilson
1991). Predator exclosures also suffer the problem of a
mismatch of scale between the area covered by
exclosures (several square meters) and the much larger
extents over which predators forage (Thrush 1999). This
study is quite unique in its ability to estimate depletion
on a relatively large, and ecologically relevant, spatial
scale without experimental artifacts.
The arms race between predators and prey
Natural selection by selective removal of prey can
have a profound influence on prey behavior, morphol-
ogy, and physiology (Genovart et al. 2010, Benkman et
al. 2013, Vedder et al. 2014). Under the selection
pressures imposed by predators, prey continuously
evolve behavioral, morphological, and physiological
defense mechanisms (Dawkins and Krebs 1979). In the
case of bivalves, they can reduce predation risk either by
burrowing deeper into the sediment (Zwarts and
Wanink 1989), building armor (Vermeij 1987), or
quickly attaining a refuge in size (Paine 1976). Cockle
burying depth is limited by their short siphons, and they
are found within a few centimeters of the surface
(Zwarts and Wanink 1989). This excludes the option
to reduce risk via burrowing deeper, and hence cockles
need to rely on predator defenses such as fast growth
and/or shell thickness (armor). Cockles longer than 16
mm cannot be ingested by Red Knots (Zwarts and
Blomert 1992, Piersma et al. 1993) and thus attain a
refuge in size (Paine 1976). Indeed, we found that Red
Knots mainly foraged on juvenile cockles of 7 mm in
length.
Cockles that survived predation by Red Knots had
heavier shells, indicating that Red Knots selectively fed
upon cockles with a light shell. Alternatively, the
observed increase in shell mass might have been an
induced predator response (Harvell 1990). Indeed, the
intraspecific competitive release due to Red Knot
predation could have accelerated a predator-induced
increase in shell mass. Nevertheless, given published
shell accumulation rates (e.g., Smith and Jennings 2000),
the magnitude of the observed increase within two weeks
in shell mass, with 21.5 percentage points, seems too
large to be accounted for by a plastic predator-induced
response alone. Furthermore, due to their digestive
constraint, Red Knots are expected to selectively feedupon on cockles with little shell mass and large flesh
mass, thus maximizing their energy intake rates (vanGils et al. 2003a). Our data does suggest that Red Knots
are capable of selecting those individuals with little shellmass and large flesh mass (Table 2).
The strength of natural selection
Estimates of natural selection gradients on morpho-
logical traits are common, but few are available forbody compositional traits (Kingsolver et al. 2012).
Compared to the standardized selection gradientsreported in the literature (Lande and Arnold 1983,
Endler 1986), the ones we found in the present studyrank among the highest (Siepielski et al. 2009, King-
solver and Diamond 2011b). For example, .99% of thelinear selection gradients reported in literature are
smaller than the ones we found on cockle length andshell mass in the predation plots (Kingsolver and
Diamond 2011a). The fact that we observed such strong(nonlinear) selection gradients implies that individual
cockles have the potential to increase fitness quitesubstantially. That this has not happened on thepopulation level (assuming that the traits have a
heritable component, e.g., Luttikhuizen et al. [2003]),suggests that cockles are limited in their evolutionary
response by, for instance, trade-offs between fitnesscomponents, or temporal and spatial fluctuations in
natural selection (Kingsolver and Diamond 2011b,Kingsolver et al. 2012).
We have investigated survival without taking repro-duction into account. Perhaps, increased survival from
predation (investing in armor) comes at the cost ofreproduction and competitive ability, thus reducing total
fitness. Interestingly, and perhaps indicative of a trade-off between investing in armor or flesh mass, the
selection gradients when predators are present showthat cockles invested in armor, but when predators were
absent cockles invested more in incorporating flesh mass(Table 2). Indeed, it has been found that a large flesh
mass increases reproduction in bivalves (Honkoop et al.1999, Beukema et al. 2001). The population response toselection is an average over space and time (Siepielski et
al. 2009). As shown by the fact that only two of the threeplots were experiencing predation in this study, Red
Knot predation pressure will vary in both space andtime (Folmer et al. 2010) and thus create temporal and
spatial fluctuations in the direction and strength ofnatural selection.
ACKNOWLEDGMENTS
We thank Ewout Adriaans for transport to Griend, Dirk deBoer and Peter van Tellingen for provisioning us on Griend,and ‘Natuurmonumenten’ for allowing us access to Griend.Anne Dekinga and Jan Drent provided valuable advice on thefieldwork, and Patrick Leven helped carrying it out. LisanneDerksen, Patrick Leven, and Jeremy Smith took measure-ments of the thousands of cockles in the laboratory. DickVisser prepared the figures, Sander Holthuijsen prepared
ALLERT I. BIJLEVELD ET AL.1954 Ecology, Vol. 96, No. 7
Appendix Fig. A1 (online), and Jan Drent and TanyaCompton gave valuable comments on previous drafts. Wewould also like to thank A. Richard Palmer and Lindsey R.Leighton for their valuable feedback on the manuscript. Ourwork was supported by core funding of NIOZ to T. Piersma,a grant from the Waddenfonds to T. Piersma (‘Metawad’, WF209925) and a NWO-VIDI grant to J. A. van Gils (No.864.09.002).
LITERATURE CITED
Armonies, W. 1992. Migratory rhythms of drifting juvenilemollusks in tidal waters of the Wadden Sea. Marine EcologyProgress Series 83:197–206.
Benkman, C. W., J. W. Smith, M. Maier, L. Hansen, and M. V.Talluto. 2013. Consistency and variation in phenotypicselection exerted by a community of seed predators.Evolution 67:157–169.
Beukema, J. J., R. Dekker, K. Essink, and H. Michaelis. 2001.Synchronized reproductive success of the main bivalvespecies in the Wadden Sea: causes and consequences. MarineEcology Progress Series 211:143–155.
Beukema, J. J., K. Essink, H. Michaelis, and L. Zwarts. 1993.Year-to-year variability in the biomass of macrobenthicanimals on tidal flats of the Wadden Sea: How predictable isthis food source for birds? Netherlands Journal of SeaResearch 31:319–330.
Bijleveld, A. I., S. Twietmeyer, J. Piechocki, J. A. van Gils, andT. Piersma. 2015. Data from: Natural selection by pulsedpredation: survival of the thickest. Dryad Digital Repository.http://dx.doi.org/10.5061/dryad.bk200.
Bijleveld, A. I., J. A. van Gils, J. van der Meer, A. Dekinga, C.Kraan, H. W. van der Veer, and T. Piersma. 2012. Designinga benthic monitoring programme with multiple conflictingobjectives. Methods in Ecology and Evolution 3:526–536.
Calsbeek, R., and R. M. Cox. 2010. Experimentally assessingthe relative importance of predation and competition asagents of selection. Nature 465:613–616.
Caro, T. 2005. Antipredator defenses in birds and mammals.University of Chicago Press, Chicago, Illinois, USA.
Cliff, A. D., and J. K. Ord. 1981. Spatial processes: models andapplications. Pion Limited, London, UK.
Compton, T. J., et al. 2013. Distinctly variable mudscapes:distribution gradients of intertidal macrofauna across theDutch Wadden Sea. Journal of Sea Research 82:103–116.
Cressie, N. A. C. 1993. Statistics for spatial data. Revisededition. John Wiley and Sons, New York, New York, USA.
Dawkins, R., and J. R. Krebs. 1979. Arms races between andwithin species. Proceedings of the Royal Society B 205:489–511.
De Montaudouin, X., and G. Bachelet. 1996. Experimentalevidence of complex interactions between biotic and abioticfactors in the dynamics of an intertidal population of thebivalve Cerastoderma edule. Oceanologica Acta 19:449–463.
de Roos, A. M., and L. Persson. 2013. Population andcommunity ecology of ontogenetic development. PrincetonUniversity Press, Princeton, New Jersey, USA.
Endler, J. A. 1986. Natural selection in the wild. PrincetonUniversity Press, Princeton, New Jersey, USA.
Flach, E. C. 1996. The influence of the cockle, Cerastodermaedule, on the macrozoobenthic community of tidal flats in theWadden Sea. Marine Ecology 17:87–98.
Folmer, E. O., H. Olff, and T. Piersma. 2010. How well do fooddistributions predict spatial distributions of shorebirds withdifferent degrees of self-organization? Journal of AnimalEcology 79:747–756.
Genovart, M., N. Negre, G. Tavecchia, A. Bistuer, L. Parpal,and D. Oro. 2010. The young, the weak and the sick:evidence of natural selection by predation. PLoS ONE 5:e9774.
Guillemette, M., A. Reed, and J. H. Himmelman. 1996.Availability and consumption of food by common eiders
wintering in the Gulf of St Lawrence: evidence of preydepletion. Canadian Journal of Zoology 74:32–38.
Gurevitch, J., J. A. Morrison, and L. V. Hedges. 2000. Theinteraction between competition and predation: a meta-analysis of field experiments. American Naturalist 155:435–453.
Harvell, C. D. 1990. The ecology and evolution of inducibledefenses. Quarterly Review of Biology 65:323–340.
Honkoop, P. J. C., J. van der Meer, J. J. Beukema, and D.Kwast. 1999. Reproductive investment in the intertidalbivalve Macoma balthica. Journal of Sea Research 41:203–212.
Jacoby, W. G. 2000. Loess: a nonparametric, graphical tool fordepicting relationships between variables. Electoral Studies19:577–613.
Jensen, K. T. 1993. Density-dependent growth in cockles(Cerastoderma edule): evidence from interannual compari-sons. Journal of the Marine Biological Association of theUnited Kingdom 73:333–342.
Johnson, D., K. Grorud-Colvert, T. Rankin, and S. Sponaugle.2012. Measuring selective mortality from otoliths and similarstructures: a practical guide for describing multivariateselection from cross-sectional data. Marine Ecology ProgressSeries 471:151–163.
Kamermans, P. 1993. Food limitation in cockles (Cerastodermaedule (L.)): influences of location on tidal flat and of nearbypresence of mussel beds. Netherlands Journal of SeaResearch 31:71–81.
Kingsolver, J. G., and S. E. Diamond. 2011a. Data from:Phenotypic selection in natural populations: What limitsdirectional selection? Dryad Data Repository. http://dx.doi.org/10.5061/dryad.7996
Kingsolver, J. G., and S. E. Diamond. 2011b. Phenotypicselection in natural populations: What limits directionalselection? American Naturalist 177:346–357.
Kingsolver, J. G., S. E. Diamond, A. M. Siepielski, and S. M.Carlson. 2012. Synthetic analyses of phenotypic selection innatural populations: lessons, limitations and future direc-tions. Evolutionary Ecology 26:1101–1118.
Kraan, C., G. Aarts, J. van der Meer, and T. Piersma. 2010.The role of environmental variables in structuring landscape-scale species distributions in seafloor habitats. Ecology 91:1583–1590.
Kraan, C., J. van der Meer, A. Dekinga, and T. Piersma. 2009.Patchiness of macrobenthic invertebrates in homogenizedintertidal habitats: hidden spatial structure at a landscapescale. Marine Ecology Progress Series 383:211–224.
Krebs, J. R., and N. B. Davies. 1997. Behavioural ecology: anevolutionary approach. Fourth edition. Blackwell, Oxford,UK.
Lande, R., and S. J. Arnold. 1983. The measurement ofselection on correlated characters. Evolution 37:1210–1226.
Levinton, J. 1972. Stability and trophic structure in deposit-feeding and suspension-feeding communities. AmericanNaturalist 106:472–486.
Luttikhuizen, P. C., J. Drent, W. van Delden, and T. Piersma.2003. Spatially structured genetic variation in a broadcastspawning bivalve: quantitative vs. molecular traits. Journalof Evolutionary Biology 16:260–272.
Mouritsen, K. N. 2004. Intertidal facilitation and indirecteffects: causes and consequences of crawling in the NewZealand cockle. Marine Ecology Progress Series 271:207–220.
Paine, R. T. 1976. Size-limited predation: an observational andexperimental approach with the Mytilus-Pisaster interaction.Ecology 57:858–873.
Parada, J. M., and J. Molares. 2008. Natural mortality of thecockle Cerastoderma edule (L.) from the Ria of Arousa (NWSpain) intertidal zone. Revista de Biologia Marina y Ocean-ografia 43:501–511.
July 2015 1955NATURAL SELECTION BY PULSED PREDATION
Peterson, C. H., and S. V. Andre. 1980. An experimentalanalysis of interspecific competition among marine filterfeeders in a soft-sediment environment. Ecology 61:129–139.
Piersma, T. 2007. Using the power of comparison to explainhabitat use and migration strategies of shorebirds worldwide.Journal of Ornithology 148:S45–S59.
Piersma, T. 2012. What is habitat quality? Dissecting a researchportfolio on shorebirds. Pages 383–407 in R. J. Fuller, editor.Birds and habitat: relationships in changing landscapes.Cambridge University Press, Cambridge, UK.
Piersma, T., R. Hoekstra, A. Dekinga, A. Koolhaas, P. Wolf,P. F. Battley, and P. Wiersma. 1993. Scale and intensity ofintertidal habitat use by knots Calidris canutus in the westernWadden Sea in relation to food, friends and foes. Nether-lands Journal of Sea Research 31:331–357.
Piersma, T., J. van Gils, P. de Goeij, and J. van der Meer. 1995.Holling’s functional response model as a tool to link thefood-finding mechanism of a probing shorebird with itsspatial distribution. Journal of Animal Ecology 64:493–504.
R Development Core Team. 2013. R: a language andenvironment for statistical computing. R Foundation forStatistical Computing, Vienna, Austria.
Rothery, P. 1988. A cautionary note on data transformation:bias in back-transformed means. Bird Study 35:219–221.
Siepielski, A. M., J. D. DiBattista, and S. M. Carlson. 2009. It’sabout time: the temporal dynamics of phenotypic selection inthe wild. Ecology Letters 12:1261–1276.
Sih, A., P. Crowley, M. McPeek, J. Petranka, and K.Strohmeier. 1985. Predation, competition, and prey commu-nities: a review of field experiments. Annual Review ofEcology and Systematics 16:269–311.
Smith, L. D., and J. A. Jennings. 2000. Induced defensiveresponses by the bivalve Mytilus edulis to predators withdifferent attack modes. Marine Biology 136:461–469.
Sutherland, W. J. 1982. Spatial variation in the predation ofcockles by oystercatchers at Traeth Melynog, Anglesey. I.The cockle population. Journal of Animal Ecology 51:481–489.
Szekely, T., and Z. Bamberger. 1992. Predation of waders(Charadrii ) on prey populations: an exclosure experiment.Journal of Animal Ecology 61:447–456.
Thrush, S. F. 1999. Complex role of predators in structuringsoft-sediment macrobenthic communities: implications ofchanges in spatial scale for experimental studies. AustralianJournal of Ecology 24:344–354.
van der Meer, J., and M. F. Leopold. 1995. Assessing thepopulation-size of the European storm-petrel (Hydrobatespelagicus) using spatial autocorrelation between counts from
segments of crisscross ship transects. ICES Journal of MarineScience 52:809–818.
van Gils, J. A., S. R. de Rooij, J. van Belle, J. van der Meer, A.Dekinga, T. Piersma, and R. Drent. 2005a. Digestivebottleneck affects foraging decisions in red knots Calidriscanutus. I. Prey choice. Journal of Animal Ecology 74:105–119.
van Gils, J. A., A. Dekinga, B. Spaans, W. K. Vahl, and T.Piersma. 2005b. Digestive bottleneck affects foraging deci-sions in red knots Calidris canutus. II. Patch choice andlength of working day. Journal of Animal Ecology 74:120–130.
van Gils, J. A., T. Piersma, A. Dekinga, and M. W. Dietz.2003a. Cost-benefit analysis of mollusc-eating in a shorebirdII. optimizing gizzard size in the face of seasonal demands.Journal of Experimental Biology 206:3369–3380.
van Gils, J. A., I. W. Schenk, O. Bos, and T. Piersma. 2003b.Incompletely informed shorebirds that face a digestiveconstraint maximize net energy gain when exploiting patches.American Naturalist 161:777–793.
van Gils, J. A., M. van der Geest, E. J. Jansen, L. L. Govers, J.de Fouw, and T. Piersma. 2012. Trophic cascade induced bymolluscivore predator alters pore-water biogeochemistry viacompetitive release of prey. Ecology 93:1143–1152.
Vedder, O., S. Bouwhuis, and B. C. Sheldon. 2014. Thecontribution of an avian top predator to selection in preyspecies. Journal of Animal Ecology 83:99–106.
Verhulst, P. H. 1838. Notice sur la loi que la populationpoursuit dans son accroissement. Correspondances Mathe-matique et Physique 10:113–121.
Vermeij, G. J. 1987. Evolution and escalation: an ecologicalhistory of life. Princeton University Press, Princeton, NewJersey, USA.
Wade, M. J., and S. Kalisz. 1990. The causes of naturalselection. Evolution 44:1947–1955.
Wilson, W. H. 1991. Competition and predation in marine soft-sediment communities. Annual Review of Ecology andSystematics 21:221–241.
Zwarts, L. 1991. Seasonal-variation in body-weight of thebivalves Macoma balthica, Scrobicularia plana, Mya arenariaand Cerastoderma edule in the Dutch Wadden Sea. Nether-lands Journal of Sea Research 28:231–245.
Zwarts, L., and A. M. Blomert. 1992. Why knot Calidriscanutus take medium-sized Macoma balthica when six preyspecies are available. Marine Ecology Progress Series 83:113–128.
Zwarts, L., and J. Wanink. 1989. Siphon size and burying depthin deposit-feeding and suspension-feeding benthic bivalves.Marine Biology 100:227–240.
SUPPLEMENTAL MATERIAL
Ecological Archives
The Appendix is available online: http://dx.doi.org/10.1890/14-1845.1.sm
ALLERT I. BIJLEVELD ET AL.1956 Ecology, Vol. 96, No. 7